Motivation

In many data-driven discriminative tasks, without specific priori knowledge, scientists often start with popular classifier models in machine learning. If works, bingo! A happy ending. the alorithm can be deployed into production and the researcch get published. Otherwise, continue to try other more sophisticated model. If all major models fail, we have to turn to other detection modalities, which usually mean the modality generating the current data set doesn't capture the information we need for the task.

A fundamental question beneath the above "trial and error" (Trial and error is a fundamental method of problem solving. It is characterised by repeated, varied attempts which are continued until success, or until the practicer stops trying) practice:
Does the dataset possess the statistical differences between different groups/classes? or Are the samples drawn from different distributions (generation processes)?

This reveals an implicit and often neglected pre-assessment step in the entire pipeline: the dataset classifiability analysis

Atom metrics (currently support 90+ metrics, here only list a few):

Metric 1: Bayes Error Rate

$$ BER= 1- E ({\max}_j P(Y=j|X) $$

It is the lowest possible test error rate in classification which is produced by the Bayes classifier. It is analogous to the irreducible error rate.

Because of noise (inherently stochastic), the error incurred by the oracle prediction model from the true distribution p(x,y) is the Bayes error.

To calculate BER, we can use the Gaussian Bayes model, based on the naive assumption:
The features (in this case, PCs returned from PCA in the latent space) conform to Multivariate Gaussian Distributions.
Each class corresponds to one Gaussian distribution.

The Bayes optimal decision boundary will correspond to the point where two densities are equal.

Why is this Guassian assumption reasonable for spectroscopic data?

According to the central limit theorem (CLT), if a statistic is a sum or average of repetitive measurements, it will be approximately normal under certain technical conditions regardless of the distribution of the individual samples. Each wave number in Raman or tof in ms, from physical process, is a an accumulation of measurement of photons/ions/particles. Features are uncorreleted. In micro level, physical process, photons/ions/particles of different frequency/energy/mass/electric charge don't interfere with each other. The coocurrence / correlation between features is a macro relation. i.e., Peak patterns.

Optional: check feature normality, e.g., by Q-Q plot

Implementation

To calculate BER, we can either use:

1. We need to integrate over the density that is not the highest one for each point. As there are two densities, we need to sum up the two integrals. We can use numerical integration package, e.g. scipy.integrate

2. use Gaussian NB classifier, e.g. sklearn.naive_bayes.GaussianNB

predict_proba() returns the probability of the samples for each class in the model

Metric2: Classification Accuracy

Choose SVM (support vector machine) as the base model and use K-fold CV (cross-validation) to pick up the best model.
CV ensures the classifier has a proper generalization capability (neither underfit or overfit)

Metric 3: info-gain

Information gain has been used in decision tree. For a specific feature, Information gain (IG) measures how much “information” a feature gives us about the class.

$IG(Y|X) = H(Y) - H(Y|X) $

In information theory, IG answers "if we transmit Y, how many bits can be saved if both sender and receiver know X?" Or "how much information of Y is implied in X?"

Attribute/feature X with a high IG is a good split on Y.

Implementation

It can be proven that Info Gain = Mutual information
We can use mutual_info_classif to calculate info gain.

Be cautious about the discrete_features parameter : {‘auto’, bool, array_like}, default ‘auto’

If bool, then determines whether to consider all features discrete or continuous. If array, then it should be either a boolean mask with shape (n_features,) or array with indices of discrete features. If ‘auto’, it is assigned to False for dense X and to True for sparse X.  

For continous feature, use discrete_features = False.

Notes

The term “discrete features” is used instead of naming them “categorical”, because it describes the essence more accurately. For example, pixel intensities of an image are discrete features (but hardly categorical) and you will get better results if mark them as such. Also note, that treating a continuous variable as discrete and vice versa will usually give incorrect results, so be attentive about that.
True mutual information can’t be negative. If its estimate turns out to be negative, it is replaced by zero.

Load Dataset

古井贡酒不同年份拉曼光谱 Raman spectra of Gujing Tribute Liquor of different ages

价格与年份基本正比，以500ml瓶装为例， 5-year 200RMB, 8-year 300RMB, 16-year 600RMB, 26-year 1700RMB

About Gujing Tribute Liquor

With a long history, Gujing Tribute Liquor with fragrant taste is one of the eight most famous liquors in China. In 196AD, Cao Cao presented the "Jiuyun Spring Liquor" that was produced in his hometown as the royal liquor, as well as its brewing methods to the Emperor Xian of Han Dynasty. During the Wan Li Reign of Ming Dynasty, it was presented to the royal court as a "tribute" all the way until Qing Dynasty, hence the liquor is named "Gujing Tribute Liquor" . On the basis of traditional processes, it has scientific recipes and technological innovations. It features "crystal clear, sweet and mellow like orchid, velvety and lasting after tasting" and brings a unique taste known for its sweetness, aroma and full flavor. It was awarded the gold medal of the national liquor-tasting conference for four times, and won the title of "National Famous Liquor". In March 2003, it was incorporated into the system for protecting original products. In 2005, it became national geographical iconic products, gained wide acclaim and has been popular both at home and abroad. (source: http://english.bozhou.gov.cn/content/33.html)

!pip install qsi-tk

from qsi import io
X,y,X_names,_,y_names = io.load_dataset("vintage_526", shift = 3000)

load dataset from 7344_Y5Y26.csv
X.shape (121, 2089)  y.shape (121,)

7344_Y5Y26.csv - Raman spectroscopic profiling dataset of 
5-year and 26-year Gujing Tribute vintage liquors. 

y = 0: 5-year
y = 1: 26-year

Each sample has 2088 Raman wavenumbers, ranging from 251 to 2338 cm-1.  
Three outlier samples were removed.   

--------------------
If you use this data set, please add the reference: 
[1] A unified classifiability analysis framework based on meta-learner and its application in spectroscopic profiling data [J]. Applied Intelligence, 2021, doi: 10.1007/s10489-021-02810-8

# io.scatter_plot(X, y, tags = list(range(len(X))))

Additional preprocessing steps

Feature selection via ElasticNet

1. Reduce dimension and remove unrelavent features for the classification problem
2. reduce the overfitting risk of successive classification model. Improve the generalization
3. By selecting only a few bunch of features, easier to interpret its chemcial meansings. i.e. gain a better understanding of the features and easy to notice the co-ocurrance of multiple variables (important for Raman, as one bond or molecule have muliptle peaks).

Highly recommendated for high-dimensional physio-chemical spectroscopic data, e.g., Raman and MALDI-TOF.

Dimensionality Reduction via PCA

Priori knowledge: The Raman spectrum data only contains additive or linear structures (each chemical bond or particle corresponds to several wave numbers), but no complex embedding structures. So non-linear dimensionality reduction methods, such as kernel PCA, LLE, t-SNE, do not suit.

Feature Scaling

Coefficients from LASSO or ElasticNet depend on the magnitude of each variable. It is therefore necessary to rescale, or standardize, the variables.
The result of centering the variables means that there is no longer an intercept.

Without feature scaling, the feature selection result can be quite different!

Generate Simulated Datasets

Dataset 1: mu1 = mu2

import cla.metrics

X,y = cla.metrics.mvg(
    nobs = 100, # number of observations / samples
    md = 0 # distance between means, respect to std, i.e. (mu2 - mu1) / std, or how many stds is the difference.
    )

Unable to determine R home: [WinError 2] The system cannot find the file specified

rpy2 3.X may not support Windows. ECoL metrics may not be available.

import matplotlib
matplotlib.rcParams.update({'font.size': 10})

from sklearn.decomposition import PCA
X_pca = PCA(n_components = 2).fit_transform(X)
cla.metrics.plotComponents2D(X_pca, y)

<AxesSubplot:>

Dataset 2: | mu1 - mu2 | = 1 std

X,y = cla.metrics.mvg(
    nobs = 100, # number of observations / samples
    md = 0.5 # distance between means, respect to std, i.e. (mu2 - mu1) / std, or how many stds is the difference.
    )

from sklearn.decomposition import PCA
X_pca = PCA(n_components = 2).fit_transform(X)
cla.metrics.plotComponents2D(X_pca, y)

<AxesSubplot:>

以上分别产生了类间距(md)为0和 0.5 std 的两个数据集。PCA降维后的可视化效果符合预期。

Metric Implementations

Metric1: Bayes Error Rate

Use sklearn.naive_bayes.GaussianNB
predict_proba() returns the probability of the samples for each class in the model

import matplotlib.pyplot as plt

bs = []
for NS in range(1, 40):
    b, _ = cla.metrics.BER(X,y, NSigma = NS, save_fig = '')
    bs.append(b)
    
plt.plot(bs)

[<matplotlib.lines.Line2D at 0x2cfe1925e20>]

BER随着采样范围增加而减小，在10sigma 左右达到稳定

bs = []
for m in [10,100,1000,10000,100000]:
    b,_ = cla.metrics.BER(X,y, nobs = int(m), save_fig = '')
    bs.append(b)
    
plt.plot(bs)

[<matplotlib.lines.Line2D at 0x2cfe6f23340>]

BER受随机采样点数影响不大，0.1~0.11范围

Metric2: Classification Accuracy

Use cv to get the best classifier. It returns accuracy and decision boundary vertices.

# NOTE: the CLF in pyCLAMs
dct, *_ = cla.metrics.CLF(X,y, show = True)
dct['classification.ACC']

0.59

Info Gain

We prefer Info Gain over Correlation, because:

Correlation only measures the linear relationship (Pearson's correlation) or monotonic relationship (Spearman's correlation) between two variables.

Mutual information is more general and measures the reduction of uncertainty in Y after observing X. It is the KL distance between the joint density and the product of the individual densities. So MI can measure non-monotonic relationships and other more complicated relationships.

mi,_ = cla.metrics.IG(X, y, show = True)

3. Define and Train a Unified Metric

Let's define a unified metric:
$ D = w_0 + w_1×BER + w_2×ACC + w_3×IG + … $

We use mvg() to generate datasets of different degrees of classifiability.

(Re)load the real dataset

from qsi import io
X,y,X_names,_,y_names = io.load_dataset("vintage_526", display = False) # salt

X, X_names = io.pre.x_binning(X, X_names, target_dim=0.1, flavor='max') # flavor = 'sum'
print('处理后数据维度：X.shape = ', X.shape)

load dataset from 7344_Y5Y26.csv
X.shape (121, 2089)  y.shape (121,)
7344_Y5Y26.csv - Raman spectroscopic profiling dataset of 
5-year and 26-year Gujing Tribute vintage liquors. 

y = 0: 5-year
y = 1: 26-year

Each sample has 2088 Raman wavenumbers, ranging from 251 to 2338 cm-1.  
Two outlier samples was removed.   

--------------------
If you use this data set, please add the reference: 
[1] A unified classifiability analysis framework based on meta-learner and its application in spectroscopic profiling data [J]. Applied Intelligence, 2021, doi: 10.1007/s10489-021-02810-8
处理后数据维度：X.shape =  (121, 208)

from cla.unify import calculate_atom_metrics
import numpy as np
dic = calculate_atom_metrics(mu = X.mean(axis = 0), s = X.std(axis = 0),
                             mds = np.linspace(0, 3, 4+3*4),
                             # repeat = 5, nobs = 100, 
                             show_curve = True, show_html = True) 

  0%|                                                                                           | 0/48 [00:00<?, ?it/s]R[write to console]: Loading required package: ECoL

100%|██████████████████████████████████████████████████████████████████████████████████| 48/48 [30:56<00:00, 31.62s/it]

visualize_dict()

generate_html_for_dict()

d	0.0	0.2	0.4	0.6	0.8	1.0	1.2	1.4	1.6	1.8	2.0	2.2	2.4	2.6	2.8	3.0
classification.ACC	0.932	0.975	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.Kappa	0.863	0.95	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.F1_Score	0.93	0.975	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.Jaccard	0.883	0.952	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.Precision	0.94	0.968	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.Recall	0.92	0.983	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.McNemar	0.476	0.396	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
classification.McNemar.CHI2	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf	inf
classification.CochranQ	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
classification.CochranQ.T	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
classification.CrossEntropy	0.235	0.169	0.021	0.009	0.014	0.041	0.058	0.065	0.051	0.042	0.036	0.029	0.026	0.023	0.02	0.018
classification.Mean_KLD	0.235	0.169	0.021	0.009	0.014	0.041	0.058	0.065	0.051	0.042	0.036	0.029	0.026	0.023	0.02	0.018
classification.AP	0.942	0.995	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.Brier	0.071	0.042	0.001	0.0	0.001	0.005	0.006	0.005	0.003	0.002	0.002	0.001	0.001	0.001	0.001	0.0
classification.ROC_AUC	0.948	0.995	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.PR_AUC	0.94	0.995	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
classification.BER	0.007	0.006	0.007	0.006	0.005	0.004	0.004	0.003	0.003	0.003	0.003	0.003	0.002	0.002	0.002	0.002
classification.SVM.Margin	5.359	12.935	27.626	47.547	74.555	100.34	129.178	174.038	225.507	244.66	307.292	325.335	379.776	420.714	484.931	530.5
correlation.IG.max	0.097	0.108	0.131	0.167	0.198	0.234	0.271	0.326	0.389	0.433	0.47	0.508	0.562	0.58	0.614	0.642
correlation.r.max	0.192	0.282	0.364	0.455	0.515	0.584	0.625	0.68	0.73	0.75	0.787	0.808	0.837	0.851	0.868	0.881
correlation.r.p.min	0.007	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
correlation.rho.max	0.193	0.278	0.362	0.463	0.525	0.599	0.642	0.702	0.757	0.779	0.802	0.827	0.844	0.85	0.859	0.862
correlation.rho.p.min	0.008	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
correlation.tau.max	0.158	0.228	0.296	0.379	0.43	0.491	0.526	0.575	0.62	0.637	0.656	0.677	0.691	0.696	0.703	0.705
correlation.tau.p.min	0.008	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.ES.max	0.39	0.586	0.779	1.016	1.195	1.431	1.592	1.847	2.129	2.254	2.543	2.733	3.045	3.234	3.474	3.717
test.student.min	0.007	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.student.min.log10	-2.199	-4.308	-6.97	-10.888	-14.154	-18.895	-22.309	-27.699	-33.915	-36.577	-42.675	-46.381	-52.8	-56.457	-61.002	-65.474
test.student.T.max	2.647	1.274	0.257	0.074	-0.008	-1.107	0.801	-0.713	0.127	-0.798	0.049	0.569	0.541	0.122	-0.015	-0.291
test.ANOVA.min	0.007	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.ANOVA.min.log10	-2.2	-4.309	-6.971	-10.888	-14.205	-18.942	-22.309	-27.805	-33.915	-36.625	-42.773	-46.711	-53.02	-56.724	-61.294	-65.741
test.ANOVA.F.max	7.627	17.261	30.495	51.75	71.406	102.472	126.72	170.709	227.302	254.138	323.477	373.73	464.177	523.2	603.549	691.646
test.MANOVA	0.329	0.244	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.MANOVA.log10	-0.808	-0.764	-3.932	-12.044	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654	-15.654
test.MANOVA.F	1.892	1.781	9.492	17.409	32.705	48.02	72.471	89.9	143.84	141.064	198.101	267.998	283.932	378.748	425.698	412.287
test.MWW.min	0.008	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.MWW.min.log10	-2.197	-4.084	-6.504	-10.179	-12.879	-16.556	-18.87	-22.411	-25.922	-27.347	-28.94	-30.704	-31.987	-32.372	-33.033	-33.266
test.MWW.U.min	4014.0	3392.667	2910.0	2328.0	1969.667	1539.667	1293.333	945.333	628.0	503.667	370.0	226.333	124.667	94.333	43.0	25.0
test.KS.min	0.008	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.KS.min.log10	-2.149	-4.121	-5.76	-9.501	-11.811	-15.967	-17.366	-23.805	-28.144	-31.137	-35.083	-38.335	-41.508	-45.376	-49.252	-52.58
test.KS.D.max	0.237	0.317	0.37	0.467	0.517	0.593	0.617	0.71	0.763	0.797	0.837	0.867	0.893	0.923	0.95	0.97
test.CHISQ.min	0.474	0.268	0.15	0.078	0.029	0.016	0.008	0.003	0.001	0.001	0.0	0.0	0.0	0.0	0.0	0.0
test.CHISQ.min.log10	-0.324	-0.575	-0.835	-1.111	-1.535	-1.805	-2.086	-2.586	-2.862	-3.219	-3.573	-3.931	-4.278	-4.67	-5.15	-5.219
test.CHISQ.CHI2.max	0.513	1.238	2.115	3.121	4.756	5.84	6.988	9.076	10.243	11.768	13.288	14.837	16.347	18.062	20.174	20.475
test.KW.min	0.008	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
test.KW.min.log10	-2.198	-4.086	-6.507	-10.182	-12.883	-16.561	-18.875	-22.416	-25.928	-27.353	-28.946	-30.711	-31.993	-32.379	-33.039	-33.273
test.KW.H.max	7.462	15.522	26.187	42.643	54.839	71.518	82.042	98.173	114.196	120.702	127.983	136.049	141.914	143.677	146.698	147.765
test.Median.min	0.0	0.0	0.007	0.12	1.353	1.3	1.993	0.067	0.167	0.873	0.867	0.873	0.66	0.127	1.033	0.34
test.Median.min.log10	-inf	-inf	-inf	-inf	-0.369	0.025	0.056	-inf	-inf	-0.687	-inf	-0.687	-0.427	-1.063	-0.033	-1.136
test.Median.CH2.max	1.0	1.0	0.963	0.781	0.403	0.307	0.286	0.853	0.827	0.56	0.508	0.56	0.515	0.743	0.335	0.699
overlapping.F1.mean	0.058	0.105	0.192	0.286	0.368	0.445	0.515	0.584	0.638	0.689	0.724	0.76	0.785	0.805	0.821	0.832
overlapping.F1.sd	0.042	0.061	0.068	0.07	0.068	0.063	0.059	0.06	0.061	0.058	0.058	0.056	0.056	0.057	0.055	0.054
overlapping.F1v.mean	0.713	0.7	0.693	0.692	0.696	0.709	0.731	0.746	0.75	0.75	0.75	0.75	0.75	0.75	0.75	0.75
overlapping.F1v.sd	0.013	0.009	0.005	0.006	0.012	0.022	0.022	0.012	0.004	0.003	0.003	0.002	0.002	0.003	0.003	0.003
overlapping.F2.mean	1.0	0.995	0.955	0.342	0.067	0.043	0.168	0.128	0.172	0.208	0.128	0.163	0.152	0.175	0.32	0.282
overlapping.F2.sd	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
overlapping.F3.mean	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
overlapping.F3.sd	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
overlapping.F4.mean	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
overlapping.F4.sd	0.001	0.001	0.001	0.001	0.002	0.001	0.001	0.0	0.001	0.0	0.001	0.001	0.001	0.001	0.001	0.001
neighborhood.N1	0.009	0.02	0.01	0.013	0.025	0.016	0.02	0.001	0.011	0.006	0.01	0.007	0.012	0.012	0.015	0.014
neighborhood.N2.mean	0.504	0.389	0.141	0.008	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005	0.005
neighborhood.N2.sd	0.501	0.487	0.348	0.09	0.071	0.071	0.071	0.071	0.071	0.071	0.071	0.071	0.071	0.071	0.071	0.071
neighborhood.N3.mean	0.207	0.207	0.203	0.199	0.194	0.19	0.183	0.178	0.172	0.168	0.162	0.158	0.154	0.152	0.146	0.143
neighborhood.N3.sd	0.011	0.011	0.011	0.01	0.01	0.01	0.01	0.011	0.012	0.012	0.013	0.014	0.015	0.016	0.017	0.019
neighborhood.N4.mean	0.508	0.387	0.132	0.003	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
neighborhood.N4.sd	0.501	0.488	0.339	0.047	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
neighborhood.T1.mean	0.002	0.002	0.002	0.003	0.001	0.001	0.002	0.001	0.001	0.002	0.003	0.002	0.003	0.001	0.003	0.002
neighborhood.T1.sd	0.028	0.033	0.022	0.035	0.015	0.008	0.033	0.012	0.007	0.032	0.047	0.027	0.04	0.009	0.049	0.025
neighborhood.LSC	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04	1.04
linearity.L1.mean	0.157	0.153	0.157	0.145	0.142	0.128	0.123	0.11	0.1	0.087	0.085	0.078	0.07	0.067	0.06	0.057
linearity.L1.sd	0.151	0.147	0.151	0.139	0.136	0.123	0.119	0.106	0.096	0.083	0.082	0.075	0.067	0.064	0.058	0.054

import pickle
import joblib

joblib.dump(dic, 'vintage_526_atom_metrics.pkl')

['vintage_526_atom_metrics.pkl']

import pickle
import joblib

dic = joblib.load('vintage_526_atom_metrics.pkl')

记录计算得到的atom metrics

train meta-learner

from cla.unify import train_metalearner, calculate_unified_metric
model = train_metalearner(M, dic['d'], cutoff = 2)

Score:  1.0
Coef and Intercept:  [[-3.47298416e-08  3.12323602e-02  2.42807027e-05  2.73138141e-05
   2.16828499e-05  1.77481776e-05  1.63564450e-04 -3.29321761e-03
  -3.32046375e-03  3.57354730e-02 -2.05716484e-04  2.79688743e-02
  -1.13816207e-03 -5.65481074e-02 -2.29551345e-03  2.96614339e-05
  -2.16302991e-04  9.02065961e-04 -1.13832441e-03  5.14304884e-03
   2.69952584e-05  1.03266718e-05 -2.64035099e-07  6.79523532e-07
  -2.06381078e-07 -1.98443344e-07]] [1.3719713e-05]

analyze between-class and in-class classifiability using meta.logistic

from cla.unify import calculate_unified_metric
calculate_unified_metric(X, y, model, keys, method = 'meta.logistic')

KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).

Exception ignored from cffi callback <function _consolewrite_ex at 0x000001725F2DE940>:
Traceback (most recent call last):
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\callbacks.py", line 133, in _consolewrite_ex
    s = conversion._cchar_to_str_with_maxlen(buf, n, _CCHAR_ENCODING)
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\conversion.py", line 138, in _cchar_to_str_with_maxlen
    s = ffi.string(c, maxlen).decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd4 in position 0: invalid continuation byte

KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  0 , in-class unified metric =  1.0493593448343555e-05
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  1 , in-class unified metric =  2.127618655446456e-06

(1.0, [1.0493593448343555e-05, 2.127618655446456e-06])

---

Try different unify strategies

This is an all-in-one example. It reuses a pickle data, and doesn't use filter

from qsi import io
X,y,X_names,_,y_names = io.load_dataset("vintage_526", display = False) # salt

X, X_names = io.pre.x_binning(X, X_names, target_dim=0.1, flavor='max') # flavor = 'sum'
print('处理后数据维度：X.shape = ', X.shape)

load dataset from 7344_Y5Y26.csv
X.shape (121, 2089)  y.shape (121,)
7344_Y5Y26.csv - Raman spectroscopic profiling dataset of 
5-year and 26-year Gujing Tribute vintage liquors. 

y = 0: 5-year
y = 1: 26-year

Each sample has 2088 Raman wavenumbers, ranging from 251 to 2338 cm-1.  
Two outlier samples was removed.   

--------------------
If you use this data set, please add the reference: 
[1] A unified classifiability analysis framework based on meta-learner and its application in spectroscopic profiling data [J]. Applied Intelligence, 2021, doi: 10.1007/s10489-021-02810-8
处理后数据维度：X.shape =  (121, 208)

meta.linear

from cla.unify import analyze
analyze(X,y, use_filter = True, method = 'meta.linear',  pkl = '20221126231414.365863.pkl')

Load atom metrics from 20221126231414.365863.pkl
before filter

Metrics above the threshold (0.5):  ['classification.CrossEntropy' 'classification.Mean_KLD'
 'classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.ANOVA.min.log10'
 'test.ANOVA.F.max' 'test.MANOVA.F' 'test.MWW.min.log10' 'test.MWW.U.min'
 'test.KS.min.log10' 'test.KS.D.max' 'test.CHISQ.min.log10'
 'test.CHISQ.CHI2.max' 'test.KW.min.log10' 'test.KW.H.max'
 'overlapping.F1.mean' 'overlapping.F1v.mean' 'neighborhood.N3.mean'
 'neighborhood.N3.sd' 'linearity.L1.mean' 'linearity.L1.sd']
Coef and Intercept:  [-8.99684453e+00 -8.99684452e+00 -5.18337153e+00  2.37537900e-03
 -2.41506876e+00  2.72898899e+01  4.89177112e+00  4.00408723e+00
 -3.28383202e+00 -2.28564022e-02 -3.70104037e-02  3.20868682e-03
 -3.51051937e-04  1.36222052e+01 -5.82097617e-03 -6.70669494e-02
 -8.80223782e+00  2.68495633e+00  6.82368943e-01  1.35767182e+01
  5.84758747e+00 -1.84780560e+01  2.20749327e+01 -2.88257925e+01
 -5.09165177e+00  6.41524845e+00  6.16850812e+00] 27.077690979573084
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).

Exception ignored from cffi callback <function _consolewrite_ex at 0x0000015E9F557B80>:
Traceback (most recent call last):
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\callbacks.py", line 133, in _consolewrite_ex
    s = conversion._cchar_to_str_with_maxlen(buf, n, _CCHAR_ENCODING)
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\conversion.py", line 138, in _cchar_to_str_with_maxlen
    s = ffi.string(c, maxlen).decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd4 in position 0: invalid continuation byte

KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  0 , in-class unified metric =  [-4.36287921]
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  1 , in-class unified metric =  [-13.78564454]

(array([1.20565795e+307]),
 [array([-4.36287921]), array([-13.78564454])],
 '20221126231414.365863.pkl')

meta.logistic

from cla.unify import analyze
analyze(X,y, use_filter = True, method = 'meta.logistic',  pkl = '20221126231414.365863.pkl')

Unable to determine R home: [WinError 2] The system cannot find the file specified

rpy2 3.X may not support Windows. ECoL metrics may not be available.
Load atom metrics from 20221126231414.365863.pkl
before filter

Metrics above the threshold (0.5):  ['classification.CrossEntropy' 'classification.Mean_KLD'
 'classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.ANOVA.min.log10'
 'test.ANOVA.F.max' 'test.MANOVA.F' 'test.MWW.min.log10' 'test.MWW.U.min'
 'test.KS.min.log10' 'test.KS.D.max' 'test.CHISQ.min.log10'
 'test.CHISQ.CHI2.max' 'test.KW.min.log10' 'test.KW.H.max'
 'overlapping.F1.mean' 'overlapping.F1v.mean' 'neighborhood.N3.mean'
 'neighborhood.N3.sd' 'linearity.L1.mean' 'linearity.L1.sd']
Score:  1.0
Coef and Intercept:  [[-5.73373270e-07 -5.73373270e-07  7.81742374e-09  1.93720782e-02
   2.72407895e-05  3.57117946e-05  3.56104936e-05  2.91484452e-05
   1.49638818e-04 -2.79887334e-03 -2.78816641e-03  2.58263316e-02
   1.98688303e-02 -1.53056212e-03 -3.23749432e-02 -2.18890719e-03
   3.90938098e-05 -2.23973768e-04  8.90254823e-04 -1.53082977e-03
   6.84549976e-03  3.60751853e-05  2.60570471e-05  4.37587961e-06
   6.39158385e-07  3.28793927e-07  3.16148007e-07]] [3.46445846e-05]
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).

Exception ignored from cffi callback <function _consolewrite_ex at 0x00000246E4556940>:
Traceback (most recent call last):
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\callbacks.py", line 133, in _consolewrite_ex
    s = conversion._cchar_to_str_with_maxlen(buf, n, _CCHAR_ENCODING)
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\conversion.py", line 138, in _cchar_to_str_with_maxlen
    s = ffi.string(c, maxlen).decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd4 in position 0: invalid continuation byte

KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  0 , in-class unified metric =  5.412587631243967e-05
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  1 , in-class unified metric =  0.001343643563679917

(1.0,
 [5.412587631243967e-05, 0.001343643563679917],
 '20221126231414.365863.pkl')

decompose.pca

analyze(X,y, use_filter = True, method = 'decompose.pca',  pkl = '20221126231414.365863.pkl')

Load atom metrics from 20221126231414.365863.pkl
before filter

Metrics above the threshold (0.5):  ['classification.CrossEntropy' 'classification.Mean_KLD'
 'classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.ANOVA.min.log10'
 'test.ANOVA.F.max' 'test.MANOVA.F' 'test.MWW.min.log10' 'test.MWW.U.min'
 'test.KS.min.log10' 'test.KS.D.max' 'test.CHISQ.min.log10'
 'test.CHISQ.CHI2.max' 'test.KW.min.log10' 'test.KW.H.max'
 'overlapping.F1.mean' 'overlapping.F1v.mean' 'neighborhood.N3.mean'
 'neighborhood.N3.sd' 'linearity.L1.mean' 'linearity.L1.sd']

Explained Variance Ratios for the first three PCs [8.18354850e-01 1.80890834e-01 5.36143576e-04]
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  0 , in-class unified metric =  520.9017735032552
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  1 , in-class unified metric =  512.5762879171997
before scaling:  -2.507408392083686e+306 [520.9017735032552, 512.5762879171997]
PC1 range:  -1911.3099069259176 3166.1374349557695
after scaling:  1.0 [0.52097747 0.52261717]

(1.0, array([0.52097747, 0.52261717]), '20221126231414.365863.pkl')

decompose.lda

analyze(X,y, use_filter = True, method = 'decompose.lda',  pkl = '20221126231414.365863.pkl')

Load atom metrics from 20221126231414.365863.pkl
before filter

Metrics above the threshold (0.5):  ['classification.CrossEntropy' 'classification.Mean_KLD'
 'classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.ANOVA.min.log10'
 'test.ANOVA.F.max' 'test.MANOVA.F' 'test.MWW.min.log10' 'test.MWW.U.min'
 'test.KS.min.log10' 'test.KS.D.max' 'test.CHISQ.min.log10'
 'test.CHISQ.CHI2.max' 'test.KW.min.log10' 'test.KW.H.max'
 'overlapping.F1.mean' 'overlapping.F1v.mean' 'neighborhood.N3.mean'
 'neighborhood.N3.sd' 'linearity.L1.mean' 'linearity.L1.sd']

Explained Variance Ratios for the first three PCs [1.]
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  0 , in-class unified metric =  953.7777989571432
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
KW Exception:  All numbers are identical in kruskal
MedianTest Exception:  All values are below the grand median (10.0).
c =  1 , in-class unified metric =  887.6152800386504
before scaling:  -inf [953.7777989571432, 887.6152800386504]
PC1 range:  -17.22896802736844 33.769284795474746
after scaling:  1.0 [0. 0.]

(1.0, array([0., 0.]), '20221126231414.365863.pkl')

Conclusion: The between-class classifiability metric is much bigger than the in-class value.

---

the salt dataset

from qsi import io
X,y,X_names,_,y_names = io.load_dataset("salt", x_range=list(range(400,1400)))
X, X_names = io.pre.x_binning(X, X_names, target_dim=0.1, flavor='max') # flavor = 'sum'
print('处理后数据维度：X.shape = ', X.shape)

load dataset from 7545.csv
X.shape (125, 1000)  y.shape (125,)

食盐拉曼光谱

["jinzhihaiyan(zhongyan)：精制海盐（中盐集团）加碘盐",
"shenjinkuangyan(zhongyan)：深井矿盐（中盐集团）加碘盐",
"aozhoutianranhaiyan(huaiyan)：澳洲天然海盐（淮盐）未加碘",
"aozhouxueyan(huaiyan)：澳洲雪盐（淮盐）未加碘"]

We use the 2nd and 3rd classes. 1 - well salt (China), 2 - sea salt (Austrilia)

仪器名称：  激光拉曼光谱  Laser Raman Spectrometer      
生产厂家：  美国 Enwave Optronics U.S.A.                
仪器型号：  Prott-ezRaman-d3        
测试参数：  激光波长 785 nm
           激光功率 450 mW Max模式 
           CCD -85℃
           积分时间30s
处理后数据维度：X.shape =  (125, 100)

Try different unify strategies

from cla.unify import analyze
analyze(X,y, use_filter = True, pkl = '20221127191822.217812.pkl', method = 'meta.linear')

Load atom metrics from 20221127191822.217812.pkl
before filter

Metrics above the threshold (0.5):  ['classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.student.T.max'
 'test.ANOVA.min.log10' 'test.ANOVA.F.max' 'test.MANOVA.F'
 'test.MWW.min.log10' 'test.MWW.U.min' 'test.KS.min.log10' 'test.KS.D.max'
 'test.CHISQ.min.log10' 'test.CHISQ.CHI2.max' 'test.KW.min.log10'
 'test.KW.H.max' 'test.Median.min' 'test.Median.min.log10'
 'overlapping.F1.mean' 'overlapping.F1.sd' 'overlapping.F1v.mean'
 'neighborhood.N3.mean' 'neighborhood.N3.sd' 'linearity.L1.mean'
 'linearity.L1.sd']
Coef and Intercept:  [ 2.76916359e-01  1.60863973e-03 -5.91476791e-02 -3.66022732e+00
 -1.24847056e+00 -1.02191720e+00  1.70679775e+00 -9.41727982e-02
 -1.82252403e-01  1.11443118e-01 -3.03151515e-03 -2.07351170e-04
 -3.43927288e+00  1.10905381e-03 -1.06687646e-02  8.06387989e-01
  4.64748456e+00  1.04938891e+00 -3.42747178e+00 -1.47822201e+00
 -8.88982547e-03  9.25551581e-01  2.31739150e-01  3.04947082e+00
 -2.05232016e+00  1.84737218e+00  1.18738669e+00  2.37477337e+00] -10.168545362724334
c =  1 , in-class unified metric =  [-4.83674827]
c =  2 , in-class unified metric =  [-5.28345614]

(array([-3.62766367]),
 [array([-4.83674827]), array([-5.28345614])],
 '20221127191822.217812.pkl')

from cla.unify import analyze
analyze(X,y, use_filter = True, pkl = '20221127191822.217812.pkl', method = 'meta.logistic')

Load atom metrics from 20221127191822.217812.pkl
before filter

Metrics above the threshold (0.5):  ['classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.student.T.max'
 'test.ANOVA.min.log10' 'test.ANOVA.F.max' 'test.MANOVA.F'
 'test.MWW.min.log10' 'test.MWW.U.min' 'test.KS.min.log10' 'test.KS.D.max'
 'test.CHISQ.min.log10' 'test.CHISQ.CHI2.max' 'test.KW.min.log10'
 'test.KW.H.max' 'test.Median.min' 'test.Median.min.log10'
 'overlapping.F1.mean' 'overlapping.F1.sd' 'overlapping.F1v.mean'
 'neighborhood.N3.mean' 'neighborhood.N3.sd' 'linearity.L1.mean'
 'linearity.L1.sd']
Score:  1.0
Coef and Intercept:  [[ 1.35800829e-08  1.79080650e-02  2.27082498e-05  2.68941122e-05
   2.76123428e-05  2.26016767e-05  1.11499735e-04 -2.06499142e-03
  -5.61227832e-04 -2.06887367e-03  1.91052481e-02  1.35331494e-02
  -1.19610027e-03 -2.73930910e-02 -1.89001196e-03  3.23139517e-05
  -1.80724310e-04  7.22407390e-04 -1.19630776e-03  5.35223776e-03
   3.71357695e-03  2.72674190e-05  3.00809899e-08  1.98636069e-05
   3.30494068e-06  5.05136354e-07  1.85420103e-06  3.70840205e-06]] [2.64054056e-05]
c =  1 , in-class unified metric =  0.0001808874752863014
c =  2 , in-class unified metric =  0.0007227458921617193

(4.341332031428809e-13,
 [0.0001808874752863014, 0.0007227458921617193],
 '20221127191822.217812.pkl')

from cla.unify import analyze
analyze(X,y, use_filter = True, pkl = '20221127191822.217812.pkl', method = 'decompose.pca')

Load atom metrics from 20221127191822.217812.pkl
before filter

Metrics above the threshold (0.5):  ['classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.student.T.max'
 'test.ANOVA.min.log10' 'test.ANOVA.F.max' 'test.MANOVA.F'
 'test.MWW.min.log10' 'test.MWW.U.min' 'test.KS.min.log10' 'test.KS.D.max'
 'test.CHISQ.min.log10' 'test.CHISQ.CHI2.max' 'test.KW.min.log10'
 'test.KW.H.max' 'test.Median.min' 'test.Median.min.log10'
 'overlapping.F1.mean' 'overlapping.F1.sd' 'overlapping.F1v.mean'
 'neighborhood.N3.mean' 'neighborhood.N3.sd' 'linearity.L1.mean'
 'linearity.L1.sd']

Explained Variance Ratios for the first three PCs [8.26779438e-01 1.72092940e-01 6.23388247e-04]
c =  1 , in-class unified metric =  518.018235293348
c =  2 , in-class unified metric =  529.3232315241899
before scaling:  1078.5798570230465 [518.018235293348, 529.3232315241899]
PC1 range:  -1952.4088786029863 3117.744458116934
after scaling:  0.4021899271419477 [0.51275101 0.51052129]

(0.4021899271419477,
 array([0.51275101, 0.51052129]),
 '20221127191822.217812.pkl')

from cla.unify import analyze
analyze(X,y, use_filter = True, pkl = '20221127191822.217812.pkl', method = 'decompose.lda')

Unable to determine R home: [WinError 2] The system cannot find the file specified

rpy2 3.X may not support Windows. ECoL metrics may not be available.
Load atom metrics from 20221127191822.217812.pkl
before filter

Metrics above the threshold (0.5):  ['classification.BER' 'classification.SVM.Margin' 'correlation.IG.max'
 'correlation.r.max' 'correlation.rho.max' 'correlation.tau.max'
 'test.ES.max' 'test.student.min.log10' 'test.student.T.max'
 'test.ANOVA.min.log10' 'test.ANOVA.F.max' 'test.MANOVA.F'
 'test.MWW.min.log10' 'test.MWW.U.min' 'test.KS.min.log10' 'test.KS.D.max'
 'test.CHISQ.min.log10' 'test.CHISQ.CHI2.max' 'test.KW.min.log10'
 'test.KW.H.max' 'test.Median.min' 'test.Median.min.log10'
 'overlapping.F1.mean' 'overlapping.F1.sd' 'overlapping.F1v.mean'
 'neighborhood.N3.mean' 'neighborhood.N3.sd' 'linearity.L1.mean'
 'linearity.L1.sd']

Explained Variance Ratios for the first component 1.0

Exception ignored from cffi callback <function _consolewrite_ex at 0x000002902A859AF0>:
Traceback (most recent call last):
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\callbacks.py", line 133, in _consolewrite_ex
    s = conversion._cchar_to_str_with_maxlen(buf, n, _CCHAR_ENCODING)
  File "C:\Users\eleve\anaconda3\lib\site-packages\rpy2\rinterface_lib\conversion.py", line 138, in _cchar_to_str_with_maxlen
    s = ffi.string(c, maxlen).decode(encoding)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xd4 in position 0: invalid continuation byte

c =  1 , in-class unified metric =  -118.84413572036078
c =  2 , in-class unified metric =  -129.31694369214372
before scaling:  -56.81520419251131 [-118.84413572036078, -129.31694369214372]
C1 range:  -11.518618553463384 11.1203427310966
after scaling:  1.0 [1. 1.]

(1.0, array([1., 1.]), '20221127191822.217812.pkl')

相比于前一个数据集，这个数据集可分性较差

Data Science Package Version Info

import sklearn
import scipy
import statsmodels
import seaborn
import numpy
import pandas
import qsi
import clams

print(sklearn.__version__, 
      scipy.__version__,
      statsmodels.__version__, 
      seaborn.__version__, 
      numpy.__version__,
      pandas.__version__,
     qsi.__version__,
     clams.__version__)

1.0.2 1.7.3 0.13.2 0.11.2 1.21.5 1.4.2